15

Is there a way to do it through the command line? man tcp tells me that I need to set tcp_nodelay=1, but I am unable to create the tcp_nodelay file under /proc/sys/net/ipv4. Please let me know if there's any way of disabling Nagle in Linux.

Jason Marks
  • 247
  • 1
  • 4
  • 11

1 Answers1

40

This flag (TCP_NODELAY) is an option that can be enabled on a per-socket basis and is applied when you create a TCP socket. This is done for a purpose: Nagle's algorithm is generally useful and helps handle network congestion. I doubt you want to disable it system-wide since your system will probably suffer from this deactivation.

To disable it for a given socket, you can apply the option TCP_NODELAY as explained here and here in C:

int flag = 1;
int result = setsockopt(sock,            /* socket affected */
                        IPPROTO_TCP,     /* set option at TCP level */
                        TCP_NODELAY,     /* name of option */
                        (char *) &flag,  /* the cast is historical cruft */
                        sizeof(int));    /* length of option value */
 if (result < 0)
    ... handle the error ...

You may have to adapt to your programming language, but basically it sets the TCP_NODELAY flag option to the socket sock, effectively disabling Nagle's algorithm. This is valid on any OS with sockets supporting the TCP standard.

If you still want to disable Nagle's algorithm system-wide, two options are available. First, you could recompile your kernel using the according flag (see your distribution manual for this). The second option is to create a software that sets the TCP_NODELAY flag on every existing connection, similar to this code. The latter option should be executed each time a new TCP connection is created on the system.

Something a bit cleaner would be to activate the low latency mode of TCP:

echo 1 > /proc/sys/net/ipv4/tcp_low_latency

update: tcp_low_latency was removed in kernel v4.14 and above.

This will give a hint to the TCP stack as to which decisions to make in order to lower the latency (Which I guess is what you are trying to achieve by disabling Nagle's algorithm). By default, it is set to optimize bandwidth ( "0" will be read from /proc/sys/net/ipv4/tcp_low_latency ).

user5994461
  • 5,301
  • 1
  • 36
  • 57
Soravux
  • 9,653
  • 2
  • 27
  • 25
  • Your answer is correct but why do you cast `&flag` to `char *`? `setsockopt` expects a `const void *` value pointer, so casting to a type is not required (every type can implicitly cast to `void *`): http://pubs.opengroup.org/onlinepubs/9699919799/functions/setsockopt.html Actually you even violate the C standard, as casting a pointer of type A to a pointer of type B where neither type is `void *` is actually undefined behavior by ISO-C. You have a pointer to an int but you claim it's a pointer to a `char *` (will work on almost all systems but is technically undefined) – Mecki Jul 18 '17 at 18:58
  • This code is not mine, but a copy-paste the second link as I stated. You might want to contact www.unixguide.net about this issue. I remember seeing a couple of different signatures for `setsockopt()`, some of which (like the first link I provided) use a `char *` for the `flag` parameter. I don't mind changing the answer, as long as there are references to a credible `setsockopt()` prototype and an a code example to perform the desired operation. – Soravux Jul 24 '17 at 19:52
  • agreed with Soravux : from Microsoft MSDN: int setsockopt( SOCKET s, int level, int optname, const char *optval, int optlen ); from linux man pages int setsockopt(int sockfd, int level, int optname, const void *optval, socklen_t optlen); from vxWorks socklib STATUS setsockopt ( int s, int level, int optname, char * optval, int optlen) and from IBM aix 7.1 int setsockopt (Socket, Level, OptionName, OptionValue, OptionLength) int Socket, Level, OptionName; const void * OptionValue; socklen_t OptionLength; – Pierre Apr 12 '18 at 17:38
  • 6
    Note that as of kernel v4.14 the tcp_low_latency no longer does anything. https://github.com/torvalds/linux/commit/b6690b14386698ce2c19309abad3f17656bdfaea#diff-ebf8916b6537cea0d23e1e0cd6abab61R356 – Spitfire19 Feb 18 '19 at 18:07